/*! This function computes the gravitational potential for ALL the particles.
* First, the (short-range) tree potential is computed, and then, if needed,
* the long range PM potential is added.
*/
void compute_potential(void)
{
int i;
#ifndef NOGRAVITY
long long ntot, ntotleft;
int j, k, level, sendTask, recvTask;
int ndone;
int maxfill, ngrp, place, nexport;
int *nsend, *noffset, *nsend_local, *nbuffer, *ndonelist, *numlist;
double fac;
double t0, t1, tstart, tend;
#ifndef NOMPI
MPI_Status status;
#endif
double r2;
t0 = second();
if(All.ComovingIntegrationOn)
set_softenings();
if(ThisTask == 0)
{
printf("Start computation of potential for all particles...\n");
fflush(stdout);
}
tstart = second();
if(TreeReconstructFlag)
{
if(ThisTask == 0)
printf("Tree construction.\n");
force_treebuild(NumPart);
TreeReconstructFlag = 0;
if(ThisTask == 0)
printf("Tree construction done.\n");
}
tend = second();
All.CPU_TreeConstruction += timediff(tstart, tend);
numlist = malloc(NTask * sizeof(int) * NTask);
#ifndef NOMPI
MPI_Allgather(&NumPart, 1, MPI_INT, numlist, 1, MPI_INT, GADGET_WORLD);
#else
numlist[0] = NumPart;
#endif
for(i = 0, ntot = 0; i < NTask; i++)
ntot += numlist[i];
free(numlist);
noffset = malloc(sizeof(int) * NTask); /* offsets of bunches in common list */
nbuffer = malloc(sizeof(int) * NTask);
nsend_local = malloc(sizeof(int) * NTask);
nsend = malloc(sizeof(int) * NTask * NTask);
ndonelist = malloc(sizeof(int) * NTask);
i = 0; /* beginn with this index */
ntotleft = ntot; /* particles left for all tasks together */
while(ntotleft > 0)
{
for(j = 0; j < NTask; j++)
nsend_local[j] = 0;
/* do local particles and prepare export list */
for(nexport = 0, ndone = 0; i < NumPart && nexport < All.BunchSizeForce - NTask; i++)
{
ndone++;
for(j = 0; j < NTask; j++)
Exportflag[j] = 0;
#ifndef PMGRID
force_treeevaluate_potential(i, 0);
#else
force_treeevaluate_potential_shortrange(i, 0);
#endif
for(j = 0; j < NTask; j++)
{
if(Exportflag[j])
{
for(k = 0; k < 3; k++)
GravDataGet[nexport].u.Pos[k] = P[i].Pos[k];
#ifdef UNEQUALSOFTENINGS
GravDataGet[nexport].Type = P[i].Type;
#ifdef ADAPTIVE_GRAVSOFT_FORGAS
if(P[i].Type == 0)
GravDataGet[nexport].Soft = SphP[i].Hsml;
#endif
#endif
GravDataGet[nexport].w.OldAcc = P[i].OldAcc;
GravDataIndexTable[nexport].Task = j;
GravDataIndexTable[nexport].Index = i;
GravDataIndexTable[nexport].SortIndex = nexport;
nexport++;
nsend_local[j]++;
}
}
}
qsort(GravDataIndexTable, nexport, sizeof(struct gravdata_index), grav_tree_compare_key);
for(j = 0; j < nexport; j++)
GravDataIn[j] = GravDataGet[GravDataIndexTable[j].SortIndex];
for(j = 1, noffset[0] = 0; j < NTask; j++)
noffset[j] = noffset[j - 1] + nsend_local[j - 1];
#ifndef NOMPI
MPI_Allgather(nsend_local, NTask, MPI_INT, nsend, NTask, MPI_INT, GADGET_WORLD);
#else
nsend[0] = nsend_local[0];
#endif
/* now do the particles that need to be exported */
for(level = 1; level < (1 << PTask); level++)
{
for(j = 0; j < NTask; j++)
nbuffer[j] = 0;
for(ngrp = level; ngrp < (1 << PTask); ngrp++)
{
maxfill = 0;
for(j = 0; j < NTask; j++)
{
if((j ^ ngrp) < NTask)
if(maxfill < nbuffer[j] + nsend[(j ^ ngrp) * NTask + j])
maxfill = nbuffer[j] + nsend[(j ^ ngrp) * NTask + j];
}
if(maxfill >= All.BunchSizeForce)
break;
sendTask = ThisTask;
recvTask = ThisTask ^ ngrp;
#ifndef NOMPI
if(recvTask < NTask)
{
if(nsend[ThisTask * NTask + recvTask] > 0 || nsend[recvTask * NTask + ThisTask] > 0)
{
/* get the particles */
MPI_Sendrecv(&GravDataIn[noffset[recvTask]],
nsend_local[recvTask] * sizeof(struct gravdata_in), MPI_BYTE,
recvTask, TAG_POTENTIAL_A,
&GravDataGet[nbuffer[ThisTask]],
nsend[recvTask * NTask + ThisTask] * sizeof(struct gravdata_in), MPI_BYTE,
recvTask, TAG_POTENTIAL_A, GADGET_WORLD, &status);
}
}
#endif
for(j = 0; j < NTask; j++)
if((j ^ ngrp) < NTask)
nbuffer[j] += nsend[(j ^ ngrp) * NTask + j];
}
for(j = 0; j < nbuffer[ThisTask]; j++)
{
#ifndef PMGRID
force_treeevaluate_potential(j, 1);
#else
force_treeevaluate_potential_shortrange(j, 1);
#endif
}
/* get the result */
for(j = 0; j < NTask; j++)
nbuffer[j] = 0;
for(ngrp = level; ngrp < (1 << PTask); ngrp++)
{
maxfill = 0;
for(j = 0; j < NTask; j++)
{
if((j ^ ngrp) < NTask)
if(maxfill < nbuffer[j] + nsend[(j ^ ngrp) * NTask + j])
maxfill = nbuffer[j] + nsend[(j ^ ngrp) * NTask + j];
}
if(maxfill >= All.BunchSizeForce)
break;
sendTask = ThisTask;
recvTask = ThisTask ^ ngrp;
#ifndef NOMPI
if(recvTask < NTask)
{
if(nsend[ThisTask * NTask + recvTask] > 0 || nsend[recvTask * NTask + ThisTask] > 0)
{
/* send the results */
MPI_Sendrecv(&GravDataResult[nbuffer[ThisTask]],
nsend[recvTask * NTask + ThisTask] * sizeof(struct gravdata_in),
MPI_BYTE, recvTask, TAG_POTENTIAL_B,
&GravDataOut[noffset[recvTask]],
nsend_local[recvTask] * sizeof(struct gravdata_in),
MPI_BYTE, recvTask, TAG_POTENTIAL_B, GADGET_WORLD, &status);
/* add the result to the particles */
for(j = 0; j < nsend_local[recvTask]; j++)
{
place = GravDataIndexTable[noffset[recvTask] + j].Index;
P[place].Potential += GravDataOut[j + noffset[recvTask]].u.Potential;
}
}
}
#endif
for(j = 0; j < NTask; j++)
if((j ^ ngrp) < NTask)
nbuffer[j] += nsend[(j ^ ngrp) * NTask + j];
}
level = ngrp - 1;
}
#ifndef NOMPI
MPI_Allgather(&ndone, 1, MPI_INT, ndonelist, 1, MPI_INT, GADGET_WORLD);
#else
ndonelist[0] = ndone;
#endif // NOMPI
for(j = 0; j < NTask; j++)
ntotleft -= ndonelist[j];
}
free(ndonelist);
free(nsend);
free(nsend_local);
free(nbuffer);
free(noffset);
/* add correction to exclude self-potential */
for(i = 0; i < NumPart; i++)
{
/* remove self-potential */
P[i].Potential += P[i].Mass / All.SofteningTable[P[i].Type];
if(All.ComovingIntegrationOn)
if(All.PeriodicBoundariesOn)
P[i].Potential -= 2.8372975 * pow(P[i].Mass, 2.0 / 3) *
pow(All.Omega0 * 3 * All.Hubble * All.Hubble / (8 * M_PI * All.G), 1.0 / 3);
}
/* multiply with the gravitational constant */
for(i = 0; i < NumPart; i++)
P[i].Potential *= All.G;
#ifdef PMGRID
#ifdef PERIODIC
pmpotential_periodic();
#ifdef PLACEHIGHRESREGION
i = pmpotential_nonperiodic(1);
if(i == 1) /* this is returned if a particle lied outside allowed range */
{
pm_init_regionsize();
pm_setup_nonperiodic_kernel();
i = pmpotential_nonperiodic(1); /* try again */
}
if(i == 1)
endrun(88686);
#endif
#else
i = pmpotential_nonperiodic(0);
if(i == 1) /* this is returned if a particle lied outside allowed range */
{
pm_init_regionsize();
pm_setup_nonperiodic_kernel();
i = pmpotential_nonperiodic(0); /* try again */
}
if(i == 1)
endrun(88687);
#ifdef PLACEHIGHRESREGION
i = pmpotential_nonperiodic(1);
if(i == 1) /* this is returned if a particle lied outside allowed range */
{
pm_init_regionsize();
i = pmpotential_nonperiodic(1);
}
if(i != 0)
endrun(88688);
#endif
#endif
#endif
if(All.ComovingIntegrationOn)
{
#ifndef PERIODIC
fac = -0.5 * All.Omega0 * All.Hubble * All.Hubble;
for(i = 0; i < NumPart; i++)
{
for(k = 0, r2 = 0; k < 3; k++)
r2 += P[i].Pos[k] * P[i].Pos[k];
P[i].Potential += fac * r2;
}
#endif
}
else
{
fac = -0.5 * All.OmegaLambda * All.Hubble * All.Hubble;
if(fac != 0)
{
for(i = 0; i < NumPart; i++)
{
for(k = 0, r2 = 0; k < 3; k++)
r2 += P[i].Pos[k] * P[i].Pos[k];
P[i].Potential += fac * r2;
}
}
}
if(ThisTask == 0)
{
printf("potential done.\n");
fflush(stdout);
}
t1 = second();
All.CPU_Potential += timediff(t0, t1);
#else
for(i = 0; i < NumPart; i++)
P[i].Potential = 0;
#endif
}
/*! This function computes the gravitational potential for ALL the particles.
* First, the (short-range) tree potential is computed, and then, if needed,
* the long range PM potential is added.
*/
void compute_potential(void)
{
int i;
#ifndef NOGRAVITY
int j, k, ret, sendTask, recvTask;
int ndone, ndone_flag, dummy;
int ngrp, place, nexport, nimport;
double fac;
MPI_Status status;
double r2;
if(All.ComovingIntegrationOn)
set_softenings();
if(ThisTask == 0)
{
printf("Start computation of potential for all particles...\n");
fflush(stdout);
}
CPU_Step[CPU_MISC] += measure_time();
if(TreeReconstructFlag)
{
if(ThisTask == 0)
printf("Tree construction.\n");
CPU_Step[CPU_MISC] += measure_time();
#if defined(SFR) || defined(BLACK_HOLES)
rearrange_particle_sequence();
#endif
force_treebuild(NumPart, NULL);
CPU_Step[CPU_TREEBUILD] += measure_time();
TreeReconstructFlag = 0;
if(ThisTask == 0)
printf("Tree construction done.\n");
}
/* allocate buffers to arrange communication */
All.BunchSize =
(int) ((All.BufferSize * 1024 * 1024) / (sizeof(struct data_index) + sizeof(struct data_nodelist) +
sizeof(struct gravdata_in) + sizeof(struct potdata_out) +
sizemax(sizeof(struct gravdata_in),
sizeof(struct potdata_out))));
DataIndexTable = (struct data_index *) mymalloc(All.BunchSize * sizeof(struct data_index));
DataNodeList = (struct data_nodelist *) mymalloc(All.BunchSize * sizeof(struct data_nodelist));
for(i = 0; i < NumPart; i++)
if(P[i].Ti_current != All.Ti_Current)
drift_particle(i, All.Ti_Current);
i = 0; /* beginn with this index */
do
{
for(j = 0; j < NTask; j++)
{
Send_count[j] = 0;
Exportflag[j] = -1;
}
/* do local particles and prepare export list */
for(nexport = 0; i < NumPart; i++)
{
#ifndef PMGRID
ret = force_treeevaluate_potential(i, 0, &nexport, Send_count);
#else
ret = force_treeevaluate_potential_shortrange(i, 0, &nexport, Send_count);
#endif
if(ret < 0)
break; /* export buffer has filled up */
}
#ifdef MYSORT
mysort_dataindex(DataIndexTable, nexport, sizeof(struct data_index), data_index_compare);
#else
qsort(DataIndexTable, nexport, sizeof(struct data_index), data_index_compare);
#endif
MPI_Allgather(Send_count, NTask, MPI_INT, Sendcount_matrix, NTask, MPI_INT, MPI_COMM_WORLD);
for(j = 0, nimport = 0, Recv_offset[0] = 0, Send_offset[0] = 0; j < NTask; j++)
{
Recv_count[j] = Sendcount_matrix[j * NTask + ThisTask];
nimport += Recv_count[j];
if(j > 0)
{
Send_offset[j] = Send_offset[j - 1] + Send_count[j - 1];
Recv_offset[j] = Recv_offset[j - 1] + Recv_count[j - 1];
}
}
GravDataGet = (struct gravdata_in *) mymalloc(nimport * sizeof(struct gravdata_in));
GravDataIn = (struct gravdata_in *) mymalloc(nexport * sizeof(struct gravdata_in));
/* prepare particle data for export */
for(j = 0; j < nexport; j++)
{
place = DataIndexTable[j].Index;
for(k = 0; k < 3; k++)
GravDataIn[j].Pos[k] = P[place].Pos[k];
#ifdef UNEQUALSOFTENINGS
GravDataIn[j].Type = P[place].Type;
#ifdef ADAPTIVE_GRAVSOFT_FORGAS
if(P[place].Type == 0)
GravDataIn[j].Soft = SphP[place].Hsml;
#endif
#endif
GravDataIn[j].OldAcc = P[place].OldAcc;
for(k = 0; k < NODELISTLENGTH; k++)
GravDataIn[j].NodeList[k] = DataNodeList[DataIndexTable[j].IndexGet].NodeList[k];
}
/* exchange particle data */
for(ngrp = 1; ngrp < (1 << PTask); ngrp++)
{
sendTask = ThisTask;
recvTask = ThisTask ^ ngrp;
if(recvTask < NTask)
{
if(Send_count[recvTask] > 0 || Recv_count[recvTask] > 0)
{
/* get the particles */
MPI_Sendrecv(&GravDataIn[Send_offset[recvTask]],
Send_count[recvTask] * sizeof(struct gravdata_in), MPI_BYTE,
recvTask, TAG_POTENTIAL_A,
&GravDataGet[Recv_offset[recvTask]],
Recv_count[recvTask] * sizeof(struct gravdata_in), MPI_BYTE,
recvTask, TAG_POTENTIAL_A, MPI_COMM_WORLD, &status);
}
}
}
myfree(GravDataIn);
PotDataResult = (struct potdata_out *) mymalloc(nimport * sizeof(struct potdata_out));
PotDataOut = (struct potdata_out *) mymalloc(nexport * sizeof(struct potdata_out));
/* now do the particles that were sent to us */
for(j = 0; j < nimport; j++)
{
#ifndef PMGRID
force_treeevaluate_potential(j, 1, &dummy, &dummy);
#else
force_treeevaluate_potential_shortrange(j, 1, &dummy, &dummy);
#endif
}
if(i >= NumPart)
ndone_flag = 1;
else
ndone_flag = 0;
MPI_Allreduce(&ndone_flag, &ndone, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
/* get the result */
for(ngrp = 1; ngrp < (1 << PTask); ngrp++)
{
sendTask = ThisTask;
recvTask = ThisTask ^ ngrp;
if(recvTask < NTask)
{
if(Send_count[recvTask] > 0 || Recv_count[recvTask] > 0)
{
/* send the results */
MPI_Sendrecv(&PotDataResult[Recv_offset[recvTask]],
Recv_count[recvTask] * sizeof(struct potdata_out),
MPI_BYTE, recvTask, TAG_POTENTIAL_B,
&PotDataOut[Send_offset[recvTask]],
Send_count[recvTask] * sizeof(struct potdata_out),
MPI_BYTE, recvTask, TAG_POTENTIAL_B, MPI_COMM_WORLD, &status);
}
}
}
/* add the results to the local particles */
for(j = 0; j < nexport; j++)
{
place = DataIndexTable[j].Index;
P[place].p.dPotential += PotDataOut[j].Potential;
}
myfree(PotDataOut);
myfree(PotDataResult);
myfree(GravDataGet);
}
while(ndone < NTask);
myfree(DataNodeList);
myfree(DataIndexTable);
/* add correction to exclude self-potential */
for(i = 0; i < NumPart; i++)
{
#ifdef FLTROUNDOFFREDUCTION
P[i].p.Potential = FLT(P[i].p.dPotential);
#endif
/* remove self-potential */
P[i].p.Potential += P[i].Mass / All.SofteningTable[P[i].Type];
if(All.ComovingIntegrationOn)
if(All.PeriodicBoundariesOn)
P[i].p.Potential -= 2.8372975 * pow(P[i].Mass, 2.0 / 3) *
pow(All.Omega0 * 3 * All.Hubble * All.Hubble / (8 * M_PI * All.G), 1.0 / 3);
}
/* multiply with the gravitational constant */
for(i = 0; i < NumPart; i++)
P[i].p.Potential *= All.G;
#ifdef PMGRID
#ifdef PERIODIC
pmpotential_periodic();
#ifdef PLACEHIGHRESREGION
i = pmpotential_nonperiodic(1);
if(i == 1) /* this is returned if a particle lied outside allowed range */
{
pm_init_regionsize();
pm_setup_nonperiodic_kernel();
i = pmpotential_nonperiodic(1); /* try again */
}
if(i == 1)
endrun(88686);
#endif
#else
i = pmpotential_nonperiodic(0);
if(i == 1) /* this is returned if a particle lied outside allowed range */
{
pm_init_regionsize();
pm_setup_nonperiodic_kernel();
i = pmpotential_nonperiodic(0); /* try again */
}
if(i == 1)
endrun(88687);
#ifdef PLACEHIGHRESREGION
i = pmpotential_nonperiodic(1);
if(i == 1) /* this is returned if a particle lied outside allowed range */
{
pm_init_regionsize();
i = pmpotential_nonperiodic(1);
}
if(i != 0)
endrun(88688);
#endif
#endif
#endif
if(All.ComovingIntegrationOn)
{
#ifndef PERIODIC
fac = -0.5 * All.Omega0 * All.Hubble * All.Hubble;
for(i = 0; i < NumPart; i++)
{
for(k = 0, r2 = 0; k < 3; k++)
r2 += P[i].Pos[k] * P[i].Pos[k];
P[i].p.Potential += fac * r2;
}
#endif
}
else
{
fac = -0.5 * All.OmegaLambda * All.Hubble * All.Hubble;
if(fac != 0)
{
for(i = 0; i < NumPart; i++)
{
for(k = 0, r2 = 0; k < 3; k++)
r2 += P[i].Pos[k] * P[i].Pos[k];
P[i].p.Potential += fac * r2;
}
}
}
if(ThisTask == 0)
{
printf("potential done.\n");
fflush(stdout);
}
#else
for(i = 0; i < NumPart; i++)
P[i].Potential = 0;
#endif
CPU_Step[CPU_POTENTIAL] += measure_time();
}
/* This function computes the gravitational potential for ALL the particles.
* It expects that the particles are predicted to the current time.
* The routine constructs a new force-tree.
*/
void compute_potential()
{
#ifdef GRAPE
compute_potential_grape();
return;
#else
double t0,t1;
int i,j;
double r2,fac;
t0=second();
if(All.ComovingIntegrationOn)
set_softenings();
printf("Start potential computation..."); fflush(stdout);
force_treebuild();
All.NumForcesSinceLastTreeConstruction = All.TreeUpdateFrequency*All.TotNumPart ; /* ensures that new tree will be constructed next time*/
for(i=1;i<=NumPart;i++)
{
force_treeevaluate_potential(i-1);
P[i].Potential += P[i].Mass/All.SofteningTable[P[i].Type]; /* removes self energy */
}
if(All.ComovingIntegrationOn)
{
fac=0.5*All.Omega0*All.Hubble*All.Hubble;
for(i=1;i<=NumPart;i++)
{
#ifdef PERIODIC
P[i].Potential = All.G*P[i].Potential;
#else
for(j=0, r2=0; j<3; j++)
r2 += P[i].PosPred[j]*P[i].PosPred[j];
P[i].Potential = All.G*P[i].Potential - fac*r2;
#endif
}
}
else
{
fac= -0.5*All.OmegaLambda*All.Hubble*All.Hubble;
for(i=1;i<=NumPart;i++)
{
P[i].Potential *= All.G;
if(fac!=0)
{
for(j=0,r2=0;j<3;j++)
r2 += P[i].PosPred[j]*P[i].PosPred[j];
P[i].Potential += fac*r2;
}
}
}
printf("done.\n"); fflush(stdout);
t1=second();
All.CPU_Potential+= timediff(t0,t1);
#endif
}
